摘要
在为债券及其他金融资产进行定价时,国债利率的期限结构是重要的参考因素,如何对其进行合理而尽量准确的预测,一直是学术研究和业务实践中的关注焦点。为了对比各种预测模型对该指标的刻画能力和预测准确程度,选取2016年1月4日至2019年12月31日的国债收益率每日收盘数据,对动态Svensson模型(简称“DS模型”)和长短时记忆神经网络模型(LSTM模型)的预测结果进行比较分析发现:DS模型和LSTM模型在样本内都有着良好的拟合效果;LSTM模型的预测效果明显好于DS模型。因此,构建基于预期收益率曲线的国债投资组合主动管理策略,并通过仿真实验证实,该策略能够实现显著高于市场基准的投资收益。
The term structure of Chinese Treasury yields is a vital reference factor for the pricing of bonds and other financial assets.Prediction of the term structure has been a hot issue in the theoretical and practical research fields.This article aims to compare the effect of different forecasting models on the interpretation and prediction of the term structure of Chinese treasury yields.It selects daily treasury yield data from January 4,2016,to December 31,2019,and uses dynamic Svensson model(DS model)and Long Short Term Memory Neural Network(LSTM)for fitting and prediction.The research results show that the DS model fits well within the sample,but the out-of-sample prediction effect is worse than the LSTM model.Based on the forecast results,this article further proposes to use the Expected Yield Curve Strategy to manage the portfolio of bonds actively.It confirms that this strategy can achieve significantly higher returns than the market benchmark.
作者
薛庆
尚雨晴
金秉臣
卢文宇
Xue Qing;Shang Yuqing;Jin Bingchen;Lu Wenyu(School of Economics and Management,China University of Petroleum Beijing Campus,Beijing 102249,China;Beijing Branch of Ningbo Bank,Beijing 100000,China)
出处
《金融理论探索》
2020年第6期32-42,共11页
Exploration of Financial Theory
基金
教育部人文社科青年基金项目“契约视角下矿权‘圈而不探’困境解决机制研究”(20YJCZH201)
中国石油大学(北京)“研究生教育质量与创新工程”项目(yjs2017031)。
